Prostate cancer is the most commonly diagnosed life-threatening cancer amongst men in many industrialised nations.1 Until the late 1980s, its incidence and mortality were rising in most countries.2 However, death rates are now falling in some parts of the world.3, 4 The greatest interest has focused on mortality trends in the United States, where a marked peak in incidence followed introduction of the prostate-specific antigen (PSA) test in 1986.4–6 Since 1991, a fall in mortality has been seen in the United States, and there has been debate over the extent to which this may be a consequence of PSA screening.6 In England and Wales, where national health policy has discouraged PSA screening, similar declines in mortality have also been observed.7 Reports of falling prostate-cancer death rates have also come from Scotland8 and Canada.9 To investigate the extent to which these patterns are mirrored in other industrialised countries, we used routinely collected data to explore recent changes in mortality trends.
Incidence and mortality from prostate cancer were rising in most countries until the late 1980s. Following a number of advances in the management of prostate cancer, including introduction of the prostate-specific antigen (PSA) test, there have been reports of declines in mortality in Canada, the United States and the United Kingdom. To investigate the extent to which this pattern was seen in other industrialised countries, we used routinely collected data to explore recent changes in prostate-cancer mortality. Trends in age-standardised death rates between 1979 and 1997 for men aged 50 to 79 years in 24 industrialised countries were compared using join point regression. Join point regression allows estimation of the annual percentage change in death rates and tests for significant changes in trend. During the period studied, age-standardised mortality increased at 1% to 2% per year in most countries. In 7 countries (Canada, United States, Austria, France, Germany, Italy and United Kingdom), a significant down-turn in age-standardised mortality was observed over the period 1988–1991. Trends in age-specific rates within these countries support a period effect on prostate-cancer mortality. Declines in mortality could result from any combination of either artefact, reduction in prostate-cancer incidence, a rise in competing causes of death or changes in the risk of death from prostate cancer. There are inconsistencies in the relationship between national mortality trends and uptake of PSA screening; further research is required to determine whether changes in death rates can be explained by international and secular variations in the treatment of prostate cancer. © 2001 Wiley-Liss, Inc.
MATERIAL AND METHODS
Mortality and population data were extracted from the World Health Organisation (WHO) Mortality Database (http://www.who.int/whosis/mort) for the period 1979–1997. Countries were included in analyses if (i) they had at least 200 deaths from prostate cancer per year, (ii) mortality data were available for at least 9 years in the period 1979–1997 and (iii) they had systems of death certification in which cause of death was assigned by a medical practitioner in ≥95% of cases and coverage was considered good.10 Deaths coded to prostate cancer using the ninth revision of the International Classification of Disease were used in analyses (ICD9-185). Age-standardised death rates were calculated for the age range 50–79 years in 5-year age bands, using the European standard population.11 Deaths in those aged over 79 years were excluded as death certification is known to be less accurate in older individuals.12 Mortality and population data from the former German Democratic Republic and the Federal Republic of Germany for the years 1979–1989 were combined to give a single rate for Germany.
Trends in age-specific and age-standardised mortality were calculated using join point regression.13 This is a form of regression analysis in which trend data can be described by a number of contiguous linear segments and “join points,” at which trends change. The parameters estimated by this model are the annual percentage change in rates, the number and location of join points and confidence intervals (CIs) for these parameters. A full description of the use of join point regression in the analysis of trends in cancer rates, with special reference to prostate cancer, is given by Kim et al.13 Analyses were conducted using Joinpoint software (version 2.5; National Cancer Institute, Bethesda, MD).
Regression models were estimated using weighted least squares, with weights proportional to the inverse of the variance of the age-standardised rates. To identify the best-fitting combination of line segments and join points, a series of permutation tests was performed, first testing Ho (no join points) vs. Ha (3 join points). Hypothesis testing proceeded sequentially, increasing the number of join points under Ho by 1 if the null hypothesis was rejected and decreasing the number of join points under the alternative hypothesis if Ho was accepted. The maximum number of join points tested was 3 in each analysis. For each model, the locations of the best-fitting join points were identified using a grid search algorithm.14 A Bonferonni correction was applied by conducting each test at the α/3 level, ensuring that the probability of a type I error (i.e., concluding that there are 1 or more join points when there are in fact none) was at most 0.05.
Figure 1 shows age-standardised mortality rates for each of the 24 countries included. There was a >5-fold difference between the lowest (Japan 1994, 15.1/100,000) and highest (Sweden 1996, 81.5/100,000) rates at the end of the time series. The lowest death rates were seen in Japan and the southern European countries (Greece, Spain, Portugal, Italy) and the highest in Scandinavia and North America. Most countries experienced a constant increase in prostate-cancer mortality over the years studied. In 7 countries (Canada, United States, Austria, France, Germany, Italy and United Kingdom), mortality trends showed a down-turn in recent years. In Belgium and Hungary, trends were static.
Estimates of the year when changes in the trend in prostate-cancer mortality occurred and of the annual percentage change in prostate-cancer mortality between these years are shown in Table I. Mortality rates increased at between 1% and 2% per year in most countries during the study period. Japan, with the lowest initial mortality rate, had one of the highest rates of increase, 2.6% per year. In countries where a down-turn in trend was identified, join points for this clustered in the time period 1988–1991, the earliest reversal in trend being seen in Italy (1988, 95% CI 1986–1991). Following the reversal in trend, the annual percentage change ranged from falls of 2.1% per year in the United Kingdom (95% CI 1.3–2.9) and Italy (95% CI 1.0–3.3) to 6% in Germany (95% CI 0.1–11.5).
|Country (range of years)||APC (95%CI)||Join point (95%CI)||APC (95%CI)||Join point(95%CI)||APC (95%CI)|
|2 join points|
|USA (1979–1987)||0.5(0.1–1.0)||1988 (1981–1990)||3.6(−1.0–8.4)||1991 (1989–1993)||−3.6 (−4.3 to −2.8)|
|UK (1979–1997)||0.1(−5.5–6.0)||1981 (1981–1992)||3.2(2.7–3.7)||1991 (1989–1995)||−2.1 (−2.9 to −1.3)|
|1 join point|
|Canada (1979–1997)||1.7(1.2–2.3)||1991 (1988–1992)||−3.0 (−4.3 to −1.6)|
|Austria (1980–1997)||1.0(0.2–1.9)||1991 (1989–1994)||−2.5 (−4.5 to −0.3)|
|France (1979–1995)||1.1 (0.6–1.6)||1989 (1987–1990)||−2.9 (−3.7 to −2.0)|
|Germany (1980–1997)||0.5 (0.3–0.8)||1995 (1993–1995)||−6.0 (−11.5 to −0.1)|
|Italy (1979–1995)||1.0 (0.1–1.7)||1988 (1986–1991)||−2.1 (−3.3 to −1.0)|
|No join point|
|Belgium (1979–1997)||−0.1 (−0.5–0.4)|
|Finland (1987–1995)||1.1 (−0.4–2.6)|
|Greece (1979–1997)||1.0 (0.5–1.4)|
|Ireland (1979–1996)||1.8 (1.0–2.7)|
|The Netherlands (1979–1995)||0.9 (0.5–1.2)|
|Norway (1986–1995)||0.4 (−0.3–1.1)|
|Portugal (1980–1997)||1.4 (1.0–1.8)|
|Spain (1980–1995)||0.3 (0.01–0.5)|
|Sweden (1987–1996)||1.3 (0.5–2.2)|
|Bulgaria (1979–1995)||0.9 (0.5–1.2)|
|Hungary (1979–1995)||−0.02 (−0.4–0.4)|
|Poland (1980–1996)||1.8 (1.4–2.1)|
|Romania (1979–1997)||1.0 (0.5–1.4)|
|Israel (1979–1996)||1.8 (0.8–2.8)|
|Japan (1979–1994)||2.6 (2.3–3.0)|
|Australia (1979–1995)||1.7 (1.3–2.1)|
|New Zealand (1979–1996)||1.1 (0.4–1.8)|
Age-specific rates for countries in which a down-turn in mortality was observed are shown in Figure 2. In Canada, United States, France, Italy and United Kingdom, a reversal in mortality trends was seen in each 5-year age group from 60 to 79 years. The timing of these changes in age-specific trends ranged from 1986 to 1992. Trends in age-specific rates were less consistent in Germany and Austria, where changes were observed mainly in the 74 to 79-year age bands.
Our analyses of recent trends in prostate-cancer mortality reveal marked differences amongst industrialised nations. In most countries, the long-term trend of rising death rates has persisted over the last 20 years; in others, there is evidence of a decline over the last decade. These declines could result from any combination of either artefact, reduction in prostate-cancer incidence, rise in competing causes of death or changes in the risk of death from prostate cancer.
There is evidence that apparent trends in prostate-cancer mortality could be artefactual. Because of its relatively slow progression and peak incidence in the elderly population, attribution of cause of death in men with prostate cancer poses particular difficulties. Changes in the interpretation of the WHO rules on assigning underlying cause of death have resulted in some spurious changes in apparent prostate-cancer mortality trends in England and Wales.3 Two U.S. studies suggest that 10% to 20% of prostate-cancer deaths may be misattributed.15, 16 In a study comparing patterns of non-prostate-cancer deaths in cohorts of men with and without prostate cancer, Newshaffer et al.16 reported that attribution of cause of death in men with prostate cancer could have been influenced by knowledge of previous treatment. Prior beliefs about the effectiveness of radical treatment may have led to a reluctance to attribute death to prostate cancer in men who had undergone “aggressive” interventions.
Autopsy rates are falling in most industrialised countries,17 and this may have a minor influence on both the detection of prostate cancer and the accuracy of death certification.8 Trends in mortality rates can also be affected by bias or lack of precision in population estimates. We restricted analyses to mortality in men under 80 years of age, which should have limited the impact of problems with validity of death certification on trends.
Changes in incidence
Trends in mortality could fall if the underlying incidence of aggressive prostate tumours was decreasing, possibly in response to changes in environmental risk factors such as diet. Such changes are hard to detect as overall estimates of incidence are heavily influenced by methods of case detection, particularly with the advent of PSA testing. However, age-specific death rates in most countries experiencing apparent falls in mortality appear consistent with a period effect, with rates declining at similar times in different age bands. Changes in chronic disease incidence resulting from changes in environmental risk factors are more commonly reflected in mortality trends as cohort effects, unless the latent time between exposure and outcome is short and all age groups experience a simultaneous change in exposure. Apart from the short-term changes in incidence driven by PSA testing, there is little evidence that prostate-cancer incidence is in decline in countries where mortality trends have reversed.2
It is also unlikely that reductions in prostate-cancer mortality are due to a real increase in the incidence of competing causes of death. However, any unrecognised adverse effects of therapy, if they induced competing causes of death such as thromboembolic disease or other malignancies, would be translated into reductions in prostate cancer-specific death rates. Life expectancy in the age range 50 to 79 years is increasing in countries experiencing a reversal in trend,10 and this is not consistent with knowledge about trends in the other major life-threatening conditions.
Increased survival: impact of health care
Earlier detection of prostate cancer following the introduction of PSA testing has distorted analyses of survival (lead-time bias) and made comparisons of recent data on survival between countries and over time very difficult.18 However, as our study was based only on the time of death, lead time bias is not an issue. Given the apparent period effects in age-specific prostate-cancer death rates, the most likely cause of the down-turns in mortality is therapeutic intervention. Two changes in the management of prostate cancer could account for this. Firstly, in countries such as the United States, the rapid rise in incidence was accompanied by an increase in the utilisation of radical surgery and radiotherapy. However, it appears unlikely that the results of radical treatments alone could have caused the changes in death rates. Using a simulation model of prostate-cancer mortality, including varying estimates of treatment efficacy and screening lead time, Etzioni et al.19 estimated the possible impact of radical treatments on death rates in the United States. Only when the effectiveness of radical treatments was set at its highest level (relative risk reduction of 0.5) and lead time at its lowest (3 years) could use of such treatments account for the observed mortality change. In the United Kingdom, the use of radical treatment has been far less but a similar down-turn in mortality has been observed.3 Secondly, management of advanced prostate cancer also developed in the 1980s with the introduction of medical anti-androgen therapies. Whilst the effectiveness of drug-induced androgen deprivation is similar to the traditional approach of surgical castration,20 increased uptake of more acceptable medical therapies could extend population survival by deferring death from prostate cancer sufficiently for competing causes to intervene. Initiation of hormone treatment earlier in the course of advanced disease might also have an influence on death rates.21 International and secular variations in health-care interventions for prostate cancer are not well described, and it is unclear whether the changes in mortality trends are matched by differences in treatment.
Impact of PSA screening
The relationship between the observed national mortality trends and the known uptake of PSA testing raises a number of paradoxes. Mortality rates are falling in the United States, where PSA screening is common,6 but not in Australia, where uptake has also been high.22, 23 In contrast, in the United Kingdom, where screening uptake is low,3 mortality rates are declining. Published reports of prostate-cancer screening in other countries are limited. PSA testing was thought to have had an impact on prostate-cancer incidence by 1988 in at least 1 area in France,24 whilst in the Italian region of Tuscany PSA was first used in the early 1990s.25 Feasibility studies for prostate-cancer screening are also ongoing in Austria26 and Italy,27 and centres in Belgium, Finland, Italy, the Netherlands and Sweden are participating in the European Randomised Study of Screening for Prostate Cancer.28 Results from the screening programme in the Tyrol region of Austria, which covers approximately 10% of the national male population and has been running since 1993, show a fall in mortality in this region but not elsewhere in the country.29 Our findings highlight the limitations of observational data in assessing the effectiveness of cancer screening, which awaits the results of ongoing randomised trials.
Our study has a number of limitations. Firstly, the data were derived from routine reporting systems, which may be of varying quality even in developed countries. However, as the focus of our study was on changes in trend, rather than comparison of rates between countries, the relevant issue is the consistency of recording within a country over time, which may be less variable. Secondly, as with any analysis of a time series, the ability to detect changes in trend is lowest at the extremes of the series and in countries with small numbers of deaths. It is possible that addition of data from more recent years may result in a statistically significant change in trend becoming apparent in other countries. Thirdly, join point regression is based on fitting trends to data using straight lines with discrete changes in rate. This clearly cannot represent the true nature of any process affecting mortality, where changes are likely to happen more gradually. However, it does provide a convenient approach to summarising patterns of changing death rates and an indication of the relative timing of changes in different countries. Finally, we examined only data on prostate-cancer mortality. Information on international and secular trends in incidence diagnostic testing and treatments, although more difficult to obtain, is required to fully interpret trends in death rates. In the absence of such detailed information, any explanation of these trends must remain speculative.
Despite these limitations, data on trends in death rates suggest that important changes have happened recently in several countries, which may be a result of health-care activity. The burden of prostate-cancer morbidity and mortality continues to rise in most parts of the world, and research into isolating the contributing causes of observed declines in mortality should have high priority.
We are grateful for comments from Drs A. Ness, R. Martin and 2 anonymous reviewers on earlier drafts of this report.